How do I request a transcription settings change?

Required Feature Flags

The following feature flags and permissions are required to use this feature:

Feature Flag	Description
Transcription	Available on contracts with transcription enabled. Settings are managed by evaluagent on your behalf.

Required Permissions:

No customer-facing permission — transcription configuration is managed by evaluagent staff

How transcription works at evaluagent

When voice recordings are imported from your contact centre platform, evaluagent transcribes them into searchable text with speaker labels and timestamps. You don't pick a vendor or wire up an external service — transcription is part of the platform.

Configuration of transcription is handled by evaluagent staff against each integration. That's deliberate: most settings need careful tuning against your audio, and getting them wrong is the difference between transcripts you can score against and transcripts that miss half the conversation.

This guide tells you what's configurable, how to request changes, and what to give us so the change is right first time.

Step 1: Decide what you want changed

These are the settings evaluagent can tune for your integrations. Pick the ones relevant to your request.

Turning transcription on or off for an integration

Whether transcription runs for a specific integration is set during provisioning. If you've added a new integration and want transcription enabled, or you need it disabled on one, ask us.

Language settings

Setting	What it does
Language mode	Auto-detect, multi-language, or single language
Primary language	The default language when running in single mode
Output locale	Regional variant, for example en-GB, en-US, en-AU
Identify languages	The set of languages to detect when running in multi mode
Low confidence action	What happens when language can't be detected — fall back to default, fail the job, or do nothing

Speaker detection

Speaker detection works out who's talking at each point in the conversation. For stereo recordings this can use channel separation. For mono, the system uses phrase libraries and algorithmic diarisation.

Setting	What it does
Detection mode	Phrase detection, agent speaks first, or customer speaks first
Assign unknown to	Default role (Agent, Customer, or Bot) for utterances that can't be confidently assigned
Agent / Customer / Bot phrases	Up to 100 phrases per role, max five words each, that indicate the speaker
Force diarisation	Apply diarisation to recordings that look stereo but actually have identical audio on both channels

For multilingual call centres, phrase libraries need to be set up per language.

Word boosting

Up to 40 domain-specific terms the transcription service should prioritise. Useful for product names, industry jargon, and words that get consistently misheard. Send us a list with examples of where each term gets misheard if you have them.

Processing options

Setting	What it does
Enhanced transcription	Higher-accuracy transcription. Incurs additional charges.
Exclude system notes	Excludes internal notes (for example Intercom internal notes) from AI processing
Volume threshold	Filters audio below a minimum volume
Speaker sensitivity	How aggressively the system detects speaker changes
Prefer current speaker	Reduces erratic speaker switching on ambiguous utterances
Remove disfluencies	Strips filler words (um, uh) from transcripts. Off by default to preserve coaching detail.

Sampling

Whether every contact gets transcribed or only a sample, and how that sample is selected. Set during provisioning. Ask us if you want it reviewed.

Setting	What it does
Sampling method	Random or filtered selection
Sampling percentage	What percentage of eligible contacts to transcribe
Minimum audio length	Skip recordings shorter than this
Lookback window	How far back to collect recordings from

Step 2: Raise the request

Email your evaluagent representative or support@evaluagent.com with:

The integration the change applies to (for example, Genesys Cloud, Aircall)
Which settings you want changed and what you want them set to
Any context that helps us tune the change — example calls, common phrases, languages spoken

If you're requesting word boosting, include the terms. If you're changing speaker detection phrases, include typical opening lines for each role and language.

Step 3: Confirm the result

After we apply the change, transcripts ingested from that point will use the new settings. Older transcripts aren't reprocessed.

Listen to a handful of new transcripts and check:

Speaker labels are landing on the right speaker
Domain terms are coming through correctly if you requested word boosting
Language is detected correctly if you changed language settings

If something's not right, reply to the same thread and we'll tune further.

What you can do yourself

Audio quality at source is on you. The cleaner the audio going in, the better the transcript coming out. The things to get right in your contact centre:

Stereo recordings where your platform supports it — far better speaker separation than mono
16kHz or higher sample rate
Clear phone lines, decent microphones, minimal background noise

evaluagent normalises audio to 16kHz during processing, so anything above that doesn't add accuracy, but anything below it loses detail.

Why I can't see transcription settings in my menu

You won't see a settings page for transcription. The configuration sits in an internal admin tool that customers don't have access to. Everything goes via your CSM or support@evaluagent.com.